Distribution Of Natural Secondary Forest Mortality In Maoershan Experimental Forest Farm Based On GWR Models

Natural secondary forest area in China occupies a large proportion in the total forest area,due to the interference of various factors,the overall stability of dynamic natural secondary forest succession is low,while the number of tree mortality affects the forest succession process,Natural secondary plays a crucial role in the process of forest dynamics,so it has the important theory and practical significance to study on spatial distribution of the number of mortality trees in natural secondary forest.Based on the data of 101 natural secondary forest permanent sample plots collected from 2004 to 2016,the global Poisson model and the GWPR model under the four spatial scales（2.5km,5km,10 km,15km）were established,while we explored the spatial distribution of the number of morality trees for the natural secondary forest in Maoer Mountain Experimental Forest Farm.The superiority and inferiority to deal with the problem of space of global model and local model under different spatial scales were analyzed.The global Poisson model was established by taken the mixed stepwise selection method to fit the relationship between the number of mortality trees and the influenced factors,finally,we found the,altitudes,slope,average DBH of stand,trees per hectare,and volume were obtained.The results showed that all the variables were statistically significant for the global Poisson model,indicating that the stand factors and topographic factors were important factors affecting the number of mortality trees for natural secondary forest,the average DBH had a negative correlation with the number of mortality,and the other four factors were positively correlated with the number of stand mortality.The average DBH of stands had the greatest influence on the number of mortality with natural secondary forest,the topographic factors such as elevation and slope have been the least effected on the number of mortality for natural secondary forest.In this paper,a local GWPR model with four spatial scales were established and the spatial distribution of the parameters of the local GWPR model under four spatial scales was plotted by ArcGIS software.The results showed that the GWPR model at each scale was consistented with the global Poisson model in the direction of the number of stand mortality,but the estimation coefficients of the GWPR model was not unique and it was a continuous change.The GWPR model coefficients can produced a localized spatial distribution in the range of 2.5km and produced the maximum range of variation.The degree of refinement of the model coefficients was better than the GWPR model at other scales.At the same time,the accuracy of the model coefficients was roughly the same at each scale.In the aspect of model fitting,the GWPR model has a high fitting accuracy,which was obviously higher than that of the global Poisson model,especially,the AIC value was significantly reduced at the small spatial scale（GWPR₂.5km）,and the model coefficients were obtained in a large range,and get better localized spatial distribution of the effect with model parameters.The GWPR model at a small scale produces a smaller range of model residuals,and the ideal spatial distribution pattern of small observation clustering with different observations was obtained.The spatial autocorrelation of the global model and the local model under four scales was analyzed by Moran I index.The results show that the model generated at a small scale（GWPR₂.5km）can effectively remove the global spatial autocorrelation of the model residuals and produced the least significant local Moran I values,with the increased of the scale,the number of significant local Moran I value were gradually increased.On the basis of reduced the effect of local spatial autocorrelation,the GWPR model（2.5km and 5km）at the smaller scale was significantly better than the GWPR model at larger scale,however,all the local model iwas still better than the global model,indicated that the local model in reduced or even eliminated the spatial autocorrelation has played a very good effect.Based on ArcGIS software,the global Poisson model and the spatial model（GWPR）of four spatial scales were used to predict the spatial distribution of the number of natural secondary forest mortality.The results showed that the prediction trend of the local model and the four regional scales for the number of secondary forest wounds was basically the same in the whole area,the number of dead strains in the northeast and southwest of the study area was more than other areas,however,there were differences in the fitting deviations of the models,the fitting error of the global Poisson model was about 6 trees,The GWPR model of the GWPR model was between 5 and 6 trees,and the GWPR model was between 5 and 6 trees,and the GWPR model of the GWPR model was between 5 and 6,The model fit was about 6 trees.